From May to November, 2019, patients followed at an outpatient eating disorders’ clinic operating as a statewide public point of reference in an university hospital were invited to take part in the study. Patients underwent a multidisciplinary treatment on weekly basis that included physicians (psychiatrists and nutrition specialist), psychologists and dietitians, scheduled according to clinical needs. Residents, students, and trainees in the areas of pediatrics, endocrinology, psychology, psychiatry and nutrition also attended the service.

Participants were informed about the objectives and the study protocols before signing the Informed Consent Form (ICF). Patients under 18 and their adult legal representative also signed the consent form previously approved by the University Ethics Committee, under the number CAAE 53641815.6.0000.5149.

AN diagnosis and recovery criteria followed the Diagnostic and Statistical Manual of Mental Disorders, 5th edition [1]. Patients with thyroid disease assessed by thyroid stimulating hormone (TSH) were excluded from the study due to changes in energy expenditure.

University students and employees composed the control group, according to the following inclusion criteria: absence of AN, female gender, BMI between 18.5 and 24.9 kg/m2, regular menstrual cycles, and absence of disease affecting energy metabolism.

Three study groups were defined: Active AN (ANact), recovered AN (ANrec) and healthy individuals (HI). Inclusion criteria in the ANact group were BMI below 18.5 kg/m2 and/or weight ad equacy below 85% of ideal weight, with the presence of

key symptoms as defined by DSM-5 (fear of gaining weight and disturbed body image). ANrec group inclusion criteria included previous AN according to the DSM-5 and BMI above 18.5 kg/m2, and/or weight adequacy above 85% of the ideal weight, maintained for a sustainable period, in addition to partial or total remission of the key symptoms as highlighted above. For this last parameter, the evaluation of the multidisciplinary team that included psychologists and psychiatrists was considered.

Nutritional assessment was carried out individually in a silent environment, at room temperature of 22 ºC to 25 ºC by the researcher and two trained assistants.

Body weight and BMI measurement were performed with a mechanical scale calibrated with ± 0.1 kg precision coupled to a stadiometer with ± 0.1 cm precision for height measurement.

Nutritional status was based on the BMI calculated according to the standard formula: weight (kg)/height (m2) [22]. Subjects were considered eutrophic when BMI values were between 18.50 kg/m2 and 24.99 kg/m2 for adults [22] or Z score ≥ − 2 and

<+1 for adolescents from 10 to 19 years old [23].

Body composition was assessed by a Quantum X® [24] bioimpedance device of low intensity (800 μA) and single frequency (50 kHz). Prior to the BIA exam, patients were instructed to undergo a food and beverage fast and to refrain from drinking alcohol and from performing physical activity. BIA was not performed during menstrual cycles. BIA provided the values for resistance (R) and reactance (Xc) by the Body Composition Program and calculation of PA was performed according to the formula: PA = arc-tangent (Xc/R) [16, 24].

BIA derived equations provided values for TBW and FFM. The formulas for FFM and TBW were: FFM (women) = − 4.104 + 0.518 x RI + 0.231 x Weight + 0.130 x Xc; TBW (women) = 0.434 x weight + 6.326 [16]. FM was calculated from FFM and TBW. FM and FFM indices were calculated from the equation: FMI = fat mass (kg)/height2 (m2); FFMI = fat free mass (kg)/height2 (m2) [24]. Normal values for FFMI are between 15.0 kg/ m2 - 16.6 kg/m2 [20].

For IC, the MetaCheck® calorimeter was employed. At a comfortable room temperature of 22 to 25ºC. All participants were instructed to fast for at least 5 hours before testing. The device estimates the individual's resting metabolic rate in kcal/day from the VO2, considering that each calorie consumed needs an amount of the supplied oxygen to be converted into energy, according to the equation by Weir [25]. Measurements were performed at the same time of the day to avoid fluctuations in weight, thus affecting body composition and RMR.

Data were analyzed with the software Statistical Package for Social Science (SPSS) version 25. Variables were classified into three categories: 1) anthropometric: BMI (kg/m2); 2) direct BIA parameters: R (Ω), Xc (Ω) and PA(o); 3) indirect BIA parameters FFM (kg), FFM (%), FM (kg), FM (%), FFMI (kg/m2), FMI(kg/m2), TBW(kg) and TBW(%).

Differences among groups (ANact, ANrec and HI) were analyzed by comparison of means for variables with normal distribution (ANOVA with post hoc’Bonferroni correction) or comparison of medians for variables with non-normal distribution (Kruskal Wallis, followed by Mann-Whitney with Bonferroni correction, in case of statistical significance). Significance threshold was set at 0.05. ROC (receiver operating characteristic) curves were constructed for the studied variables, with ANact x ANrec and ANrec x HI as comparison groups.

Considering that the three study groups presented progressive and ordered values in their anthropometric and BIA variables (Table 1), an attempt to establish correlations between the independent variables and classification in each study group was made by means of an ordinal regression. The ordered categories ANact (outcome 2), ANrec (outcome 1) or HI (outcome 0) were considered dependent variables. The variables listed in Table 1 were used as independent variables. As a first step, separate univariate regressions were performed for each variable and those with p < 0.2 were chosen to enter the models. As a second step, Spearman's correlation analysis was performed with the variables obtained in the first step in order to select groups of variables in which all correlation coefficients were below 0.7. Separate ordinal regression models were performed for each one of these groups. Model fit was tested by the Deviance chi-square test and assumption of proportional odds was tested by the test of parallel lines. BMI was not included among independent variables due to its determinant role in the definition of the 3 study groups. Ordinal regression results were presented as odds ratio.

Bivariate regression was performed to explore the correlation between PA and RMR.